lifecycle_examples

Enhancement-81 — session-lifecycle + memory audit

Interactive ngspice workflows with OSDI devices, probed for correctness and bounded memory — plus the two resource fixes this enhancement ships.

What the audit found (all healthy)

• Re-sourcing the same deck is idempotent; remcirc + a new deck resolves cleanly to the new circuit.
• The E-66 Monte-Carlo reset idiom is leak-free in practice: 100 reset+op iterations grow the ngspice program size by ~6 kB/iteration (bounded < 20 kB in the check) with the solution exact throughout.
• Plots accumulate one per analysis (the documented E-66 trap for big loops) and destroy all genuinely frees them — the numbering restarts at tran1.

The two fixes

1. ft_ckspace (the "approaching max data size" warning) now honors the same set no_mem_check opt-out as the output-size estimate in outitf.c, and warns once per excursion above the 95% threshold (re-arming below 90%) instead of repeating on every check.
2. The pre_osdi already-loaded note (E-76) now tells you what to do about it: "restart ngspice to load a recompiled file". The stale-model behavior it warns about is pinned by the suite: overwriting a loaded .osdi and re-loading the same path keeps the old model for the rest of the session — same effective behavior as the pre-E-76 silent shadowing, now with instructions.

Files

verify_lifecycle.py (9 checks, self-contained decks), lres.va (the minimal OSDI resistor the probes instantiate).